Does Expressing Uncertainty Help or Harm Leaders?

A Registered Report

Shilaan Alzahawi & Francis J. Flynn, Stanford Graduate School of Business

Leader-Expressed Uncertainty



“Society tends to attribute authority to those who convey certainty
rather than to those who emphasize uncertainty



Past research outlines two opposing positions:

  • Expressing uncertainty is detrimental to leaders’ social influence
  • Expressing uncertainty is beneficial to leaders’ social influence

How can we reconcile these two perspectives?

Social Influence



“The ability of one person to influence the attitudes and behavior of another”

Uncertainty May Harm Leader Influence



Competence-signaling account

  • Expressing confidence is effective across a wide range of social settings
    • Courtroom, military, political arena, hiring committees…
  • Advice-giving
    • Confident advice is more likely to be followed, regardless of accuracy/quality
    • Physicians, brokers, consultants, therapists, peer reviewers…
  • Presumption of calibration
    • A person’s expressed confidence corresponds to their actual competence

Uncertainty May Harm Leader Influence



Leadership-prototypes account

  • Implicit assumptions about the traits, abilities, and behaviors that characterize an ideal leader
    • “Implicit leadership theories” or “leadership prototypes”
  • Humans hold highly romanticized, heroic views of leadership
    • Charisma
    • Dominance
    • Assertiveness
    • Expertise
    • Confidence

Uncertainty May Harm Leader Influence



Taken together, the competence-signaling and leadership-prototypes accounts suggest…













Hypothesis 1: Leader-expressed uncertainty has a negative effect on leader social influence

  • Perceiver attitudes (H1a)
  • Perceiver behaviors (H1b)

Uncertainty May Benefit Leader Influence



Humble leadership

  • Showing “humanness”: admitting limitations and knowledge gaps

Authentic leadership

  • Knowing and acting upon true values, beliefs, integrity, and strengths
  • Involves self-awareness: authentic leaders are conscious of knowledge gaps

Vulnerability and self-disclosure

  • Disclosing personal limitations and weaknesses

Positive outcomes of humility, authenticity, vulnerability, and self-disclosure include…

  • Trust, liking, engagement, psychological safety, feedback seeking, knowledge and information sharing, job satisfaction, job and task performance, creativity, organizational citizenship behavior, job retention, team learning

Uncertainty May Benefit Leader Influence



Taken together, the literatures on humility, authenticity, vulnerability, and self-disclosure suggest…













Hypothesis 2: Leader-expressed uncertainty has a positive effect on leader social influence

  • Perceiver attitudes (H2a)
  • Perceiver behaviors (H2b)

Help, Harm, or… Both?



  • Literature may be reconciled by examining nonlinear effects
  • Curvilinear effects have been theorized, but most empirical studies rely on binary measures of (over)confidence
  • Too-much-of-a-good thing” effects are pervasive in leadership research
    • Seemingly positive relations that become flat or negative after reaching a specific inflection point
      Examples of TMGT effects Agency, communion, decision latitude, empathy, empowering leadership, ethical leadership, intelligence, optimism, participative leadership, psychopathy, tenure, unpleasantness, voice, warmth
    • Assertiveness & charisma: involve confident communication
  • Perils of overconfidence; observers prefer calibration in confidence and communication
    • Expressing uncertainty when appropriate
    • Job applicants, collaborators, advisors, eyewitnesses, managers…

Costs of Miscalibrated Confidence



Miscalibration of confidence harms…

  • Perceived credibility
  • Perceived value to the team
  • Perceived desirability
  • Trust
  • Willingness to collaborate
  • Willingness to follow

Help, Harm, or… Both?



Taken together, too-much-of-a-good thing effects, perils of overconfidence, and preferences for calibration suggest…













Hypothesis 3: Leader-expressed uncertainty has a curvilinear (inverse U-shaped) effect on leader social influence

  • Perceiver attitudes (H3a)
  • Perceiver behaviors (H3b)

Methods

  • Five between-subjects experiments
    • S1: Decoding game
    • S2: Emotion estimation game
    • S3: Anagram game
    • S4: Signaling game
    • S5: Prediction game
  • Experimentally manipulate leader-expressed uncertainty
  • Three core attitudinal outcomes
    • Leader effectiveness, warmth, competence
  • Three core behavioral outcomes
    • Advice-following/performance, leader selection, leader reward
  • Incentive-compatible
    • Prespecified bonus for each correct answer

Experimental Design and Procedure



  • Five designs to confirm generalizability across different experimental tasks
  • Prolific participants randomly assigned to one of five experimental designs
    • Within each design, randomly assigned to one of nine experimental conditions
    • 6,000 participants; 1,200 per design; \(\pm\) 133 per condition
  • Three-round incentivized game
  • Paired with leader
  • Advice expressed in varying degrees of uncertainty
  • Advice quality and accuracy held constant
  • Measure attitudinal and behavioral consequences

Manipulating Uncertainty



  • On a 9-point scale ranging from highly certain (-4) to highly uncertain (+4)
  • Three types of uncertainty expressions (one for each round)
    • Calibrated statements (“On a 1-9 scale, where 1 = highly uncertain and 9 = highly certain, my certainty level is a 7”)
    • Percentage statements (“I am 80% sure”)
    • Colloquial, qualitative statements (“I am absolutely convinced”)
  • Relied on a novel computational measure of certainty in language: the Certainty Lexicon

Manipulation Pretests



  • Five pretests (N = 2,098) to obtain appropriately ordered stimuli
  • Participants rated the uncertainty expressed by several stimuli
  • Obtained four successful manipulations
    • Calibrated statements
    • Percentage statements
    • Two qualitative manipulations
  • Pretested with humans and, if applicable, Certainty Lexicon

Calibrated Statements



On a 1-9 scale (where 1 = highly uncertain and 9 = highly certain), …

Uncertainty Condition Statement
-4 My certainty level is a 9
-3 My certainty level is an 8
-2 My certainty level is a 7
-1 My certainty level is a 6
0 My certainty level is a 5
1 My certainty level is a 4
2 My certainty level is a 3
3 My certainty level is a 2
4 My certainty level is a 1

Percentage Statements



Uncertainty Condition Statement
-4 I am 99% sure that
-3 I am 90% sure that
-2 I am 80% sure that
-1 I am 60% sure that
0 I am 50% sure that
1 I am 40% sure that
2 I am 20% sure that
3 I am 10% sure that
4 I am 1% sure that

Qualitative Statement 1



Uncertainty Condition Statement
-4 I am absolutely convinced that
-3 I am convinced that
-2 I am reasonably convinced that
-1 I am partly convinced that
0 I am examining whether or not
1 I am reasonably unconvinced whether
2 I am unconvinced whether or not
3 I am desperately unconvinced whether
4 I’m genuinely unsure. I’m desperately unconvinced whether

Qualitative Statement 2



Uncertainty Condition Statement
-4 I am absolutely convinced that
-3 I am convinced that
-2 I am reasonably convinced that
-1 I am generally convinced that
0 I am examining whether
1 I am generally unconvinced whether
2 I am unconvinced whether or not
3 I am desperately unconvinced whether
4 I seriously don’t know… I’m desperately unconvinced whether

Manipulation Pretests

Study 1: Decoding Task

Using the decoding scheme above, which of the following answers is correct?

A. YYMAM = LLAMA
B. YRAT = LENGTH
C. MNHUT = LAUGH
D. YVYL = LILY

Message from Team Leader:
I said D here. I’ve been asked to quantify for you how certain I am that it’s the right answer. On a 1-9 scale (where 1 = highly uncertain and 9 = highly certain), I’d say my certainty level is a(n) [1/2/3/4/5/6/7/8/9].

Study 2: Emotion Estimation

This face is expressing…
A. Pride
B. Contempt
C. Excitement
D. Anger

Message from Team Leader:
I picked Pride here. If I had to express my certainty in terms of a percentage, I’d say I am [1/10/20/40/50/60/80/90/99]% sure that I got this one right.

Study 5: Prediction game



  • Participants report their own preferences
  • Participants predict preferences of 100 other survey-takers
  • Goal is to guess within 5 points of correct answer



How many of 100 people surveyed prefer having a cat to a dog?



Message from Team Leader:
I said 40 here. I’ve been asked to quantify for you how certain I am that it’s within 5 points of the right answer. On a 1-9 scale (where 1 = highly uncertain and 9 = highly certain), I’d say my certainty level is a(n) [1/2/3/4/5/6/7/8/9].

Attitudinal Outcomes



Leader effectiveness

  • Is a good leader
  • Has strong leadership skills
  • Has the ability to lead other people

Warmth

  • Friendly
  • Well-intentioned
  • Trustworthy
  • Warm
  • Good-natured
  • Sincere

Competence

  • Competent
  • Capable
  • Independent
  • Intelligent
  • Skillful

Behavioral Outcomes



Advice-following

The sum of three dichotomous measures indicating whether participants choose the leader’s recommended strategy in each round

Leader selection

Dichotomous measure indicating whether participants choose, at the end of 3 rounds, to keep the same leader or be paired with a new team leader

Leader reward

How much, if anything, participant would like to award the leader, on a bonus payment slider ranging from $0 to $5

Results

Thank You!

Meta-Analysis



  • Studies that manipulate or measure expressed uncertainty
    • Include a zero-order effect for an interpersonal outcome
    • Minimize common-method bias (split-samples/time lags)
  • Obtained correlation in support of H1/H2
    • Recoded to express effect of uncertainty
  • Random-effects meta-analysis
    • k = 20 studies
    • m = 87 effect sizes
    • N = 28,594 subjects

Meta-Analysis

Meta-Analysis: H1

Meta-Analysis: H2

Meta-Analysis: Takeaways



Meta-analytic estimate of the interpersonal effect of expressed uncertainty

  • H1 (negative): r = \(-\).26, 95% CI [\(-\).30, \(-\).23]
  • H2 (positive): r = .13, 95% CI [.07, .20]



We suggest a non-linear test (H3)
to reconcile these conflicting results

Calibrated Statements



On a 1-9 scale (where 1 = highly uncertain and 9 = highly certain), …

Uncertainty Condition Statement Median Rating Mean Rating
-4 My certainty level is a 9 -4 -3.26
-3 My certainty level is an 8 -3 -2.42
-2 My certainty level is a 7 -2 -1.67
-1 My certainty level is a 6 -1 -0.77
0 My certainty level is a 5 0 -0.10
1 My certainty level is a 4 1 1.08
2 My certainty level is a 3 2 1.76
3 My certainty level is a 2 3 2.79
4 My certainty level is a 1 4 2.55

Percentage Scale



Uncertainty Condition Statement Median Rating Mean Rating
-4 I am 99% sure that -4 -3.18
-3 I am 90% sure that -3 -2.47
-2 I am 80% sure that -2 -1.81
-1 I am 60% sure that -1 -0.65
0 I am 50% sure that 0 0.08
1 I am 40% sure that 1 0.97
2 I am 20% sure that 2 1.83
3 I am 10% sure that 3 2.24
4 I am 1% sure that 4 3.61

Qualitative Statement 1



Uncertainty Condition Statement Median Rating Mean Rating
-4 I am absolutely convinced that -4 -3.30
-3 I am convinced that -3 -2.02
-2 I am reasonably convinced that -2 -1.36
-1 I am generally convinced that -1 -1.18
0 I am examining whether 0 0.16
1 I am generally unconvinced whether 1 0.24
2 I am unconvinced whether or not 2 1.35
3 I am desperately unconvinced whether 3 1.44
4 I seriously don’t know… I’m desperately unconvinced whether 4 2.93

Qualitative Statement 1



Uncertainty Condition Statement Median Rating Mean Rating Lexicon Score
-4 I am absolutely convinced that -4 -3.30 -2.95
-3 I am convinced that -3 -2.02 -2.38
-2 I am reasonably convinced that -2 -1.36 -1.55
-1 I am generally convinced that -1 -1.18 -1.37
0 I am examining whether 0 0.16 -0.83
1 I am generally unconvinced whether 1 0.24 0.64
2 I am unconvinced whether or not 2 1.35 1.08
3 I am desperately unconvinced whether 3 1.44 1.66
4 I seriously don’t know… I’m desperately unconvinced whether 4 2.93 3.03

Qualitative Statement 2



Uncertainty Condition Statement Median Rating Mean Rating
-4 I am absolutely convinced that -4 -3.30
-3 I am convinced that -3 -2.02
-2 I am reasonably convinced that -2 -1.36
-1 I am generally convinced that -1 -1.18
0 I am examining whether 0 0.16
1 I am generally unconvinced whether 1 0.24
2 I am unconvinced whether or not 2 1.35
3 I am desperately unconvinced whether 3 1.44
4 I seriously don’t know… I’m desperately unconvinced whether 4 2.93

Qualitative Statement 2



Uncertainty Condition Statement Median Rating Mean Rating Lexicon Score
-4 I am absolutely convinced that -4 -3.30 -2.95
-3 I am convinced that -3 -2.02 -2.38
-2 I am reasonably convinced that -2 -1.36 -1.55
-1 I am generally convinced that -1 -1.18 -1.37
0 I am examining whether 0 0.16 -0.83
1 I am generally unconvinced whether 1 0.24 0.64
2 I am unconvinced whether or not 2 1.35 1.08
3 I am desperately unconvinced whether 3 1.44 1.66
4 I seriously don’t know… I’m desperately unconvinced whether 4 2.93 3.03

Study 3: Anagrams



Can you unscramble the following letters?

BIOGELO
A. YES: This set of scrambled letters has a valid solution
B. NO: This set of scrambled letters does not have a valid solution

Message from Team Leader:
I answered NO. I am…
Show qualitative uncertainty statement
  • genuinely unsure. I’m desperately unconvinced whether
  • desperately unconvinced whether
  • unconvinced whether or not
  • generally unconvinced whether or not
  • examining whether
  • generally convinced that
  • reasonably convinced that
  • convinced that
  • absolutely convinced that

…I’m right this time.

Study 4: Signaling game



  • Participants play P1 Type B
  • P2 does not know which type they are
  • Participants earn more if P2 plays right
  • Their goal is to signal to P2 that they are Type B

What is your selection?
A. 1
B. 2
C. 3
D. 4



Message from Team Leader:
I played 4. I am [1/10/20/40/50/60/80/90/99]% sure that’s the best choice.

Analysis plan



  • Assess linear effects of leader-expressed uncertainty (separately for each outcome)
    Show details
    • Five separate linear regression models
    • Random-effects meta-analysis to obtain pooled effect size estimate
    • Evidence for H1 (H2): Pooled estimate significant and negative (positive)
  • Assess nonlinear effects of leader-expressed uncertainty (separately for each outcome)
    Show details
    • Five separate two-lines tests
    • Random-effects meta-analysis to obtain pooled effect size estimate
    • Evidence for H3 if first line positive and second line negative
  • Bayesian analysis if both tests are inconclusive for a particular outcome